Recently, the hammer new machine conference will let the University of Science and Technology Flight Voice input method fire! Everyone said: No matter how fast or complicated, this artifact can be recognized. Even a few days ago, a pop-up friend who took the movie took this magic. I have been popular in the IT industry for nearly 10 years. What is the contextual recognition? I feel that the world has turned over... In order to show that I have not been dry for years. I decided to explain briefly to the voice input method of the 啥科大讯.
It all stems from something called "deep learning." Deep learning is a neural network that establishes and simulates the human brain for analytical learning, enabling the machine to learn the law from a large amount of historical data, so as to intelligently identify new samples or predict the future to achieve the same human thinking ability.
There are three prerequisites for influencing the development of deep learning: algorithms, calculations, and data. With the development of Internet technology, the data that humans can access and use is exploding. It is estimated that the global data volume will exceed 4 trillion GB in 2020, which solves the data acquisition problem of deep learning development. In terms of algorithms, the most commonly used DNN algorithm (k-nearest neighbor classification algorithm) can better simulate the multi-layer deep transfer process of human brain neurons and solve the complex problems in intelligent speech. Then the calculation is performed. It is understood that there are roughly 100 billion neurons in the human brain, and there are about 5,000 synapses in each neuron. To make the machine infinitely close to human thinking ability means to simulate more neurons and synapses. Will bring huge computing challenges.
In order to improve the intelligent speech recognition rate, HKUST has announced the “Xinfei Super Brain Project†very early, and plans to simulate 1/10 of human brain neurons in order to make the company's intelligent speech equipment have preliminary human thinking ability. To achieve a 1/10 depth simulation of human brain neurons means that the University of Science and Technology is facing enormous challenges of thousands of times of training data and thousands of times of model parameters. Larger, more stored super-computing clusters, better deep learning parallelization and cluster scheduling algorithms, and deep-customized artificial neural network-specific chip systems have also become urgent needs of HKUST.
When I mentioned this, I had to mention the wave. Inspur started cooperation with Keda Xunfei a long time ago. The high-performance computing cluster designed and designed by Keda Xunfei uses NF5280M4 and NF5288M4 servers as cluster nodes. Each NF5280M4 server is equipped with one NVIDIA M40 accelerator card. The NF5288M4 server is equipped with four NVIDIA M40 accelerator cards. At present, these servers have been applied to many companies such as Keda Xunfei to support deep learning applications.
Live Broadcast Display Screen,Live Streaming Big Size Projection Screen,Mobile App Interaction Screen,Wireless Projection Screen
Jumei Video(Shenzhen)Co.,Ltd , https://www.jmsxdisplay.com